12 May 2016 Applying traditional signal processing techniques to social media exploitation for situational understanding
Author Affiliations +
Abstract
Signal processing techniques such as filtering, detection, estimation and frequency domain analysis have long been applied to extract information from noisy sensor data. This paper describes the exploitation of these signal processing techniques to extract information from social networks, such as Twitter and Instagram. Specifically, we view social networks as noisy sensors that report events in the physical world. We then present a data processing stack for detection, localization, tracking, and veracity analysis of reported events using social network data. We show using a controlled experiment that the behavior of social sources as information relays varies dramatically depending on context. In benign contexts, there is general agreement on events, whereas in conflict scenarios, a significant amount of collective filtering is introduced by conflicted groups, creating a large data distortion. We describe signal processing techniques that mitigate such distortion, resulting in meaningful approximations of actual ground truth, given noisy reported observations. Finally, we briefly present an implementation of the aforementioned social network data processing stack in a sensor network analysis toolkit, called Apollo. Experiences with Apollo show that our techniques are successful at identifying and tracking credible events in the physical world.
© (2016) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Tarek Abdelzaher, Heather Roy, Shiguang Wang, Prasanna Giridhar, Md. Tanvir Al Amin, Elizabeth K. Bowman, Michael A. Kolodny, "Applying traditional signal processing techniques to social media exploitation for situational understanding", Proc. SPIE 9831, Ground/Air Multisensor Interoperability, Integration, and Networking for Persistent ISR VII, 98310R (12 May 2016); doi: 10.1117/12.2229723; https://doi.org/10.1117/12.2229723
PROCEEDINGS
12 PAGES


SHARE
RELATED CONTENT


Back to Top